How to find the questions buyers ask AI assistants about your category
In short
The questions that matter are the ones a buyer asks before they know your name. Find them in your sales calls, support tickets and the reviews of your rivals, write them the way a person types, and leave your brand out. Group them by topic, because the topics where you are named, even once, are where you have a foothold.
Which questions count
Only questions without your name tell you whether you are being found. A question with your name in it gets your name back almost every time, so it says nothing about a buyer who has never heard of you.
Keep a few branded questions if you want to check what assistants say about you. Count them apart, and never add them to the figure for being found.
Ask what a stranger to your brand would ask. That is the buyer you are trying to reach.
Where to find real questions
The best questions are ones real people already asked you or someone like you. Look in these places first.
| Where | What you find |
|---|---|
| Sales calls and demos | The problem a buyer had before they found you, in their own words |
| Support tickets | What new customers expected the product to do |
| Reviews of your rivals | What people chose a rival for, and what disappointed them |
| Forums and communities | Questions asked in public, before anyone mentions a brand |
| Your own site search | What visitors could not find on your pages |
| The "People also ask" box in search | Related questions people type into search engines |
How to word them
- Write a full question, the way a person types into a chat. Write "What tool sends invoice reminders automatically?", not "invoice reminder tool".
- Leave your brand out. Leave your rivals out too, unless the question is a comparison a buyer would really ask.
- Mix two shapes. Some questions ask for a product, such as "What tool does X?". Others ask how to do something, such as "How do I do X?". Both lead to recommendations.
- Give each question a topic, one per job your product does.
- Aim for at least three questions per topic, so one odd answer cannot decide a whole topic.
Here is what a first set can look like. The product is made up: an invoicing app for freelancers.
| Question | Topic | Shape |
|---|---|---|
| How do I get clients to pay invoices on time? | Late payments | How to |
| What tool sends invoice reminders automatically? | Late payments | Asks for a tool |
| Which invoicing app handles more than one currency? | Multi-currency | Asks for a tool |
| How should a freelancer track expenses for tax? | Tax | How to |
| Is there a free invoice template that adds up totals for me? | Templates | Asks for a tool |
Why topics matter more than single questions
A single question is one reading. A topic is several, and it shows where you stand. The numbers below are made up to show the idea. They are not a measurement of any real company.
| Topic | Answers that named the app |
|---|---|
| Late payments | 3 of 12 |
| Multi-currency | 1 of 12 |
| Tax | 0 of 12 |
| Templates | 0 of 12 |
Example values only. In total the app is named in 4 of 48 answers.
That is the map to work from. The app has a foothold on late payments and almost none elsewhere. A page for a topic where you are already named sometimes is a different job from a page for a topic where you are never named.
Test the list before you trust it
- Ask each question in the assistants your buyers use, with web search on, and keep the answers.
- Drop questions every assistant answers without naming any product. They are useful to read, but they cannot be won by a recommendation.
- Ask again on another day before deciding a topic is lost. Answers change on their own.
- Keep the same questions from one month to the next. A changed question cannot be compared with the old one.
Agentcites writes a first list of up to 50 questions from your own site, each with a topic, and keeps the ones that name your brand apart. You review the list before the first scan and can add or drop questions. The buyer questions page shows how.
Questions
How do I discover the non-branded questions people ask about my niche?
Start with the words buyers already use with you: sales calls, support tickets and reviews of your rivals. Rewrite each as a full question without any brand name, and group them by the job your product does.
How many questions do I need?
Enough for at least three per topic. For example, ten topics with three questions each is 30 questions. Asked on four assistants, that gives 120 answers to read.
Should I include my competitors' names in the questions?
Only in comparisons a buyer would really ask, and count them separately. A question that names a rival invites that rival into the answer, so it says little about who an assistant would pick unprompted.
Can I see what buyers actually type into ChatGPT?
Not directly. The assistants do not publish what people ask them. That is why the best source is what buyers already said to you, rewritten as questions.